Statistical methods of unmeasured confounder control based on negative control theory / 中华流行病学杂志

ABSTRACT

Controlling unmeasured confounders in non-randomized controlled studies is challenging. Negative control theory is based on the theoretical concept that the test result of negative controls must be negative. Setting appropriate negative control incorporates the specificity of association into population studies for the identification and control of unmeasured confounders. This paper explains the principles to control unmeasured confounders using negative control theory from a statistical perspective. A detailed introduction of derived methods based on negative control theory is also introduced, including adjusted standardized mortality ratio method, calibrating P-value method, generalized difference-in-difference model and double negative control method. The reasonable application of those derived methods is also comprehensively summarized based on representative case studies. Negative control is an important statistical design to identify, revise and control unmeasured confounders and a valuable method for comparative effectiveness research based on real-world data.